{"id":"a4724f38-4fe9-456c-947e-456efa1eb433","arxiv_id":"2502.07952","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"A revenue-sharing referral fee in a Bertrand duopoly admits parameters where the retailer's payoff exceeds its single-agent profit while the seller's equilibrium price is below the retailer's, a win for both firms and consumers.","lead":"Economists analyze a modified Bertrand pricing game in which an independent seller pays the retailer a share of its revenue, and they map the Nash equilibria for every referral fee and cost level. They find a fee sweet spot where the retailer earns more than it would alone, the seller stays in the program, and consumers pay less than without the program.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The red-region price is below the retailer-alone price but above the seller's own standalone price; the headline consumer-surplus gain is baseline-dependent and Section 1.2's 'either' claim is false.","rationale":"The reader's verdict of CONDITIONAL is appropriate, but the most load-bearing issue is not primarily the equilibrium refinements. The formal construction in Sections 3-4 does identify parameters where the retailer's fee income exceeds its single-agent profit and where the seller's price is below the retailer's single-agent price, so the core existence claim is sound. However, the paper's own exposition (Section 1.2 and the conclusion) presents this as a gain over both the retailer's and the seller's independent prices, which is false. The abstract's phrase 'increase consumer surplus' is thus ambiguous: it holds relative to a retailer-only baseline but fails relative to the seller's standalone option. Since the model itself introduces an outside option for the seller in Section 6, the seller-alone counterfactual is not an extraneous reinterpretation; it is the model's own relevant alternative. This concern does not overturn the existence theorem, so REJECT is not warranted, but it strengthens the case for a conditional accept that requires the paper to state and defend the counterfactual underlying 'consumer surplus.' This is why I keep the reader's CONDITIONAL verdict unchanged while identifying a different, more direct weakness than the refinement stack.","tokens_in":38712,"tokens_out":23999,"duration_ms":207071,"concrete_test":"Recompute the full outside-option equilibrium for the Figure 5 right parameter c_r=0.9, c_s=0.83, δ=0.003: compute α^(o)_* from Eq. (28), the resulting seller price p_s(α^(o)_*), and compare it with p_r*=0.95, p_s^0=(1+c_s)/2=0.915, and p_s^(ℓ)*=(1+c_s+δ)/2=0.9165. If p_s(α^(o)_*) exceeds p_s^(ℓ)* or the retailer's equilibrium payoff falls below π_r,r(p_r*), the abstract's win-win claim fails under the seller's outside-option counterfactual. Independently, verify the Section 1.2 'either' statement by checking p_s*(α†_r)>p_s^0 for the same parameters.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The existence result itself is internally consistent, but the central 'win-win' claim is not robust to the choice of counterfactual. In the red region α∈[α_s*, α_s*,r*], the seller fulfills demand at p_s*(α), the maximizer of π_s,s(p)=((1−α)p−c_s)q(p). Because α>0, the seller's effective marginal cost is c_s/(1−α)>c_s, so for any regular demand p_s*(α) is strictly above the no-program price p_s^0=argmax_p (p−c_s)q(p). For the paper's own linear demand, p_s*(α)=(1+c_s/(1−α))/2>(1+c_s)/2=p_s^0. Thus the fact that p_s*(α)≤p_r* does not imply consumers gain relative to the independent seller's best alternative. Section 1.2 states that the price is decreased 'relative to the price either the retailer or the independent seller would have set independently,' which is false: p_s* is always greater than the seller's standalone price. When the Section 6 outside option is the relevant counterfactual, the leaving price is p_s^(ℓ)*=(1+c_s+δ)/2, and p_s*<p_s^(ℓ)* only if α<δ/(c_s+δ), a condition not guaranteed at α=α†_r. Hence the headline improvement is a transfer relative to a retailer-only world, not a welfare gain relative to the seller's own alternative.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper analyzes a two-player Bertrand game in which a retailer and an independent seller sell an identical good; if the seller fulfills demand, it remits a share α of revenue to the retailer. The authors derive the pure-strategy Nash equilibria of the 'staying' subgame (Proposition 1 and Eqs. (18)-(19)), apply admissibility and relative Pareto optimality to select outcomes (Section 4), then extend the game to an endogenous referral fee (Section 5) and to an outside option for the seller (Section 6). The headline claim is that there are parameter regions in which the seller fulfills demand at a price below the retailer's standalone price while the retailer's payoff is at least as large as its standalone profit, so that both firms and consumers can benefit.","tokens_in":38912,"tokens_out":6514,"duration_ms":60039,"significance":"The paper's equilibrium characterization is technically substantial and largely transparent: the threshold fees α_r*, α_s*, α_s*,r*, and α†_r are solved in closed form from first-order and indifference conditions (Appendix C.3), no parameters are fit to data, and the limitations (no closed form for α†_s, unsimplified p_s*/p† ordering, perfect-information assumption) are acknowledged. The no-split-market result (Proposition 1) is proved under weak assumptions, and the identification of a red region where p_s* ≤ p_r* and the retailer can earn more than its standalone profit is a genuine and falsifiable prediction. However, the welfare interpretation is considerably more fragile than the abstract suggests: the consumer-surplus gain is relative to the retailer-only counterfactual, not to the seller's standalone price, and the sharp point predictions depend on a specific stack of equilibrium refinements.","major_comments":[{"comment":"The claim that the price is decreased 'relative to the price either the retailer or the independent seller would have set independently' is false as stated. In the red region the seller fulfills demand at p_s* = argmax_p ((1−α)p − c_s)q(p); because α > 0, p_s*(α) > p_s^0 = argmax_p (p − c_s)q(p) for any regular demand. For the paper's linear demand, p_s*(α) = (1 + c_s/(1−α))/2 > (1 + c_s)/2 = p_s^0. Thus Section 4.4.4's statement that customers buy at a lower price 'than without the revenue sharing program' is valid only against the retailer-alone counterfactual, not against the seller's own standalone alternative. The abstract and Section 1.2 must be revised to name the correct counterfactual, or the paper must prove a condition such as α < δ/(c_s + δ) in the outside-option extension; at α = α†_r this inequality is not guaranteed.","section":"Section 1.2 and Section 4.4.4"},{"comment":"The sharp conclusions that the retailer always induces the seller to fulfill demand and sets the fee α* depend on a specific stack of refinements: rejecting weakly dominated strategies, relative Pareto optimality among equilibria, and an ex ante choice of continuation profile ρ. Section 4.2 concedes that a lowest-price criterion would instead select the retailer-fulfills-at-p_r* equilibrium, where the seller gets zero; and Corollary 3 shows that for general ρ the fee-optimization results are only interval bounds, with α*(ρ) = α_r* differing from α*(ρ-bar) = ᾱ. This is load-bearing for the paper's own summary that 'the retailer always prefers to allow the independent seller to fulfill demand in equilibrium.' The authors should either prove which headline results are invariant to the refinement/ρ choice or explicitly qualify those results as selection-dependent.","section":"Sections 4.2, 5.1, and 5.3; Corollary 3"},{"comment":"The outside-option extension does not currently deliver the advertised two-sided improvement in general. The seller's participation constraint is π_s^(eq) ≥ π_s^(ℓ)(δ), and α^(o)* is solved in closed form only in the special case α_max ≤ min{α†_r, α_s*} (Eq. (30)); for general ρ and δ it is left as the abstract optimization in Eq. (28). Moreover, the consumer-price comparison in Section 6 would require a condition such as p_s*(α^(o)*) < p_s^(ℓ)*(δ), which for linear demand is α^(o)* < δ/(c_s + δ); no such condition is proved. The Section 6 results should be presented as conditional existence results over specific parameter ranges rather than as a general 'often' statement about simultaneous improvements.","section":"Section 6.2 and Eq. (30)"}],"minor_comments":[{"comment":"There are several typographical errors: 'Bertand' in the Section 6.2 header, 'revenue sharking' in the first paragraph of Section 5, and 'indepedent' in Section 7; these should be corrected.","section":"Section 6.2 header and Section 5"},{"comment":"The notation ρ and ρ-bar (or ρ and ρ) is easy to confuse in the text, especially because the overline is not visually distinct in equations; renaming these to something like ρ_L and ρ_H would improve readability.","section":"Table 1 and Eq. (24)-(25)"},{"comment":"The definition of α†_s is introduced informally in Section 3.3 as the fee making p_sind ≤ p† equivalent, then defined precisely in Appendix C.4 as the largest fee satisfying π_r,s(p_sind, α†_s) = π_r,r(p_r*); the main text could state this construction at the point of introduction to avoid ambiguity.","section":"Section 3.3 and Appendix C.4"}],"recommendation":"major_revision","confidential_remarks":"The paper is a genuine theory contribution with a substantial equilibrium characterization, and the authors are unusually transparent about its soft spots. The main obstacle to acceptance is the overstated welfare claim in the abstract and Section 1.2, which is false against the seller's standalone price; this can be fixed by rewriting the claim around the retailer-only counterfactual or by adding conditions on α and δ. The dependence of the headline results on relative Pareto optimality and on the ex ante continuation profile ρ is also likely to attract referee concern, so a sensitivity analysis or an explicit statement of which results are selection-independent would materially strengthen the manuscript. The fit with econ.TH is appropriate."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"I read this one with some care, because the setup seems unassuming and the payoff coupling turns out to change the Bertrand logic in a real way. The retailer earns α p_s q_s when the seller fulfills demand, and the seller's effective marginal cost becomes c_s/(1−α). That simple twist generates a rich but tractable equilibrium set. What is actually new is the full pure-Nash classification for this game, the five threshold fees that delineate the regions, and the result that no split-market equilibrium exists. The derivations look internally consistent, and the paper is transparent about what it cannot do in closed form (α†_s, some price orderings). That is real content, and it gives platform economics a baseline that existing Bertrand, two-part tariff, and revenue-sharing papers do not provide.\n\nThe soft spots are real but localized. The big one is the consumer-surplus claim. In the red region the seller fulfills at p_s*(α), which maximizes ((1−α)p−c_s)q(p). For any α>0 this is strictly above the seller's standalone price p_s^0 = argmax (p−c_s)q(p). So the Section 1.2 statement that the price decreases relative to what either the retailer or the independent seller would set independently is false. The stress-test note is right: the win is relative to a retailer-only world, not relative to the seller's own best alternative. The paper should either weaken that claim or state the counterfactual explicitly. Also, the point predictions rest on a stack of refinements—weak dominance, relative Pareto optimality, and a continuation profile ρ. The paper is honest about this, and Section 4.2 concedes that a lowest-price criterion would select the retailer-fulfills equilibrium. So \"the retailer always prefers the seller to fulfill\" is not robust to a plausible alternative refinement. That matters for the fee-optimization conclusions, which for general ρ are only interval bounds anyway. The outside-option extension is a useful addition, though the additive δ is clearly a modeling convenience. Citation gap: Cachon and Lariviere (2005) on revenue-sharing contracts is a notable omission.\n\nWho is this for? Applied theorists working on platform pricing, marketplace referrals, and make-or-buy with shared revenue. The equilibrium classification is worth having as a baseline even if you disagree with the refinements. It deserves a serious referee: the math is coherent, the contribution is new, and the flaws are fixable claims rather than load-bearing errors. Send it out, but the authors should be pushed to correct the counterfactual and to present the refinement dependence as a feature to be examined, not an afterthought.","headline":"A genuinely new Bertrand variant with a solid equilibrium classification, but the advertised consumer-surplus win is benchmark-dependent; the paper deserves a serious referee and a careful rewrite of its claims.","tokens_in":39664,"tokens_out":1729,"would_cite":true,"duration_ms":19059,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["91A10","91B24","91A80"],"pacs":[],"model":"deepseek-v4-flash","headline":"In a shared-revenue Bertrand price game, the paper identifies cost and fee parameters under which an independent seller's optimal-price equilibrium gives consumers a lower price and gives the retailer a payoff above its own single-agent…","keywords":["Bertrand competition","revenue sharing","referral fee","Nash equilibrium","equilibrium refinement","Pareto optimality","outside option","platform pricing"],"falsifier":"For the linear demand curve $q(p)=1-p$ with costs $c_r=0.90$, $c_s=0.87$ and fee $\\alpha=0.08$ (the right subplot of Figure 2), apply the lowest-price equilibrium criterion instead of relative Pareto optimality: the refined outcome flips from the seller fulfilling at $p^\\dagger$ to the retailer fulfilling at $p_r^*$ with $p_r^*<p^\\dagger$, so the retailer's 'always prefer the seller' conclusion fails in that region. An empirical version: measure fees and prices on a real referral platform; a fee above $\\alpha_{\\max}$ with sellers still participating would contradict the outside-option model.","tokens_in":38323,"feed_emoji":"📈","tokens_out":12156,"duration_ms":102450,"temperature":0.7,"pith_summary":"The paper models a marketplace where a retailer can either sell a product itself or let an independent, lower-cost seller take the sale and pay the retailer a share $\\alpha$ of revenue. The main result is that this shared-revenue setup has a middle range of referral fees in which the independent seller sells at its own optimal price, that price is no higher than the retailer's single-agent price, and the retailer's fee income can exceed the profit it would earn selling alone. The authors solve the corresponding Bertrand price game and use equilibrium refinements to predict which outcome appears for each cost pair and fee. They then let the retailer choose the fee; in the fee-optimization game the retailer always prefers a fee that keeps the seller in the market, earns at least its single-agent profit, and stays bounded away from 1. Adding an outside option in which the seller can quit and compete directly ties the maximal fee to the seller's outside payoff, so the model explains when referral programs should exist and how large the fee should be.","feed_headline":"Revenue sharing can cut prices and raise the retailer's profit","feed_subtitle":"A shared-fee price game predicts sweet spots where independent sellers, customers, and platforms all gain.","key_machinery":"The load-bearing device is the pair of indifference prices $p_r^{\\mathrm{ind}} = c_r/(1-\\alpha)$ and $p_s^{\\mathrm{ind}} = c_s/(1-\\alpha)$: the retailer is indifferent between selling itself and collecting the fee when the seller prices at $p_r^{\\mathrm{ind}}$, while $p_s^{\\mathrm{ind}}$ is the seller's breakeven price. Around these, parameter space is organized by threshold fees $\\alpha_{s*}$, $\\alpha_{r*}$, $\\alpha_{s*,r*}$, $\\alpha^\\dagger_r$, $\\alpha^\\dagger_s$, and by the price $p^\\dagger$ defined through $\\pi_{r,s}(p^\\dagger)=\\pi_{r,r}(p_r^*)$. The mechanics are a Bertrand undercutting constraint: the seller must keep its price low enough that the retailer does not prefer to undercut and take the market, but high enough to clear its own cost; in the middle fee range the two constraints jointly admit the seller's optimal price $p_s^*$ as the refined equilibrium, which is exactly where the Pareto improvement appears.","core_discovery":"The paper's central claim is that revenue sharing can Pareto-improve on direct retailing: for costs $c_s \\le c_s^*$ and fees $\\alpha$ in the interval $[\\alpha_{s*}, \\alpha_{s*,r*}]$, the refined equilibrium has the independent seller fulfilling all demand at the seller's single-agent optimal price $p_s^*$, with $p_s^* \\le p_r^*$, so consumers pay less than they would under the retailer's own profit-maximizing price. In that same region Lemma 6 shows that at fee $\\alpha = \\alpha^\\dagger_r$ the retailer's fee income $\\pi_{r,s}(p_s^*, \\alpha^\\dagger_r)$ is at least its direct-sale profit $\\pi_{r,r}(p_r^*)$, so the program can raise the retailer's payoff while lowering the price to the customer. Section 5 extends this: after the retailer chooses the referral fee, every admissible, Pareto-optimal, subgame-perfect equilibrium gives the retailer a payoff at least $\\pi_{r,r}(p_r^*)$, with the equilibrium fee $\\alpha^*$ always bounded away from 1, and the socially suboptimal low-fee outcome never observed.","pith_inferences":["Editorial inference: the point prediction that the retailer always prefers the seller to fulfill demand rests on the relatively Pareto-optimal selection; if a platform coordinates on lowest-price equilibria, the outcome in the $p^\\dagger \\le p_s^*$ region flips to the retailer serving demand at $p_r^*$, reversing the surplus comparison there.","Editorial inference: the model gives a testable fee benchmark—estimate $p_r^*$, $p_s^*$, $c_r$, $c_s$ from demand data and the model predicts the Pareto-improving fee lies in $[\\alpha_{s*}, \\alpha_{s*,r*}]$; an observed fee outside that interval signals unobserved costs or a different selection rule.","Editorial inference: with heterogeneous sellers, a single aggregate fee acts as a participation tax; the model suggests low-cost sellers stay and high-cost sellers leave, so welfare gains depend on the cost distribution and seller-specific fees may dominate.","Editorial inference: under demand or cost uncertainty, the continuum of equilibria and the $\\rho$-dependence would become a distribution over outcomes, and the paper's interval bounds (Corollary 3, Eqs. 26-27) are the natural starting objects for that extension."],"forward_implications":["In the middle-fee region, opening a platform to an independent seller with a cost advantage strictly below $c_s^*$ lowers the consumer price and can raise the retailer's income above its own monopoly profit.","A profit-maximizing platform should not charge the highest possible referral fee; the optimal fee is bounded away from 1 and is set just low enough to keep the more efficient seller active.","When the seller's cost advantage is small ($c_s\\ge c_s^*$), the seller-optimal Pareto-improving equilibrium does not exist, so revenue sharing is only a win when the efficiency gap is large enough.","A credible outside option for the seller caps the referral fee at $\\alpha_{\\max}$ and rules out the high-fee equilibria in which the retailer ends up selling directly at $p_r^*$.","In equilibrium of the fee-optimization game, the region where the seller must price at the retailer's indifference point (low fee) is never chosen, and the retailer always weakly outperforms its single-agent payoff."],"supporting_citations":[{"why":"Defines the classic price-competition baseline that the staying and leaving subgames extend.","marker":"[Bertrand, 1883]"},{"why":"Provides the equilibrium characterization for Bertrand competition with asymmetric costs used as the leaving subgame.","marker":"[Blume, 2003]"},{"why":"Supplies the undominated Bertrand-equilibrium result underlying the leaving-subgame payoffs and outside-option threat.","marker":"[Kartik, 2011]"},{"why":"Gives the definitions of weak dominance, relative Pareto optimality, and subgame perfection used to refine the thick equilibrium set.","marker":"[Rasmusen, 1989]"},{"why":"Justifies the inadmissibility of weakly dominated strategies, pruning the continuum of Nash equilibria.","marker":"[Govindan and Wilson, 2016]"},{"why":"Formalizes the outside-option principle that the seller's leaving threat must be credible, capping the fee in Section 6.","marker":"[Watson, 2020]"}],"fun_headline_variants":["Shared revenue price game reveals win-win for sellers and buyers","Revenue-sharing Bertrand game shows Pareto gains for all","Sharing revenue beats direct selling in price game","Shared-fee game shows when outsourcing helps everyone profit","Revenue sharing can reduce prices while raising retailer profit"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The point predictions—that the retailer wants the seller to serve demand and sets the fee at $\\alpha_{r*}$—depend on selecting among a continuum of Nash equilibria by relative Pareto optimality and by fixing a continuation profile (a rule for which price is played in the remaining interval); if players instead coordinate on the lowest-price equilibrium, the retailer may serve demand itself and the fee conclusions downgrade to interval bounds.","fun_headline_variants_meta":{"raw":{"variants":["Shared revenue price game reveals win-win for sellers and buyers","Revenue-sharing Bertrand game shows Pareto gains for all","Sharing revenue beats direct selling in price game","Shared-fee game shows when outsourcing helps everyone profit","Revenue sharing can reduce prices while raising retailer profit"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000664,"raw_usage":{"total_tokens":3024,"prompt_tokens":929,"completion_tokens":2095,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":545,"completion_tokens_details":{"reasoning_tokens":2021}},"tokens_in":545,"tokens_out":2095,"duration_ms":13458,"temperature":1.0,"reasoning_tokens":2021,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-08T11:21:39.373908+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"For the linear demand curve $q(p)=1-p$ with costs $c_r=0.90$, $c_s=0.87$ and fee $\\alpha=0.08$ (the right subplot of Figure 2), apply the lowest-price equilibrium criterion instead of relative Pareto optimality: the refined outcome flips from the seller fulfilling at $p^\\dagger$ to the retailer fulfilling at $p_r^*$ with $p_r^*<p^\\dagger$, so the retailer's 'always prefer the seller' conclusion fails in that region. An empirical version: measure fees and prices on a real referral platform; a fee above $\\alpha_{\\max}$ with sellers still participating would contradict the outside-option model.","supporting_citations":[],"review_version":1}